Gender Bias and Female Labor Participation
How do gender biases among women and men impact the female labor participation in Rural Uzbekistan?
Introduction
The recent discourse on feminism reveals contrasting narratives between “Western” and “Eastern” perspectives:
“Feminist intellectual debate on an international level continues to be dominated by American and European feminism emanating from women’s experience within stable liberal democracies… They (Central Asian women) could put hard statistics on the table demonstrating the level of gender equality achieved in the communist bloc in the areas of work, education and health” (Corcoran-Nantes 2005)
The evolution of feminism in the West, marked by waves ranging from suffrage to equal pay and the contemporary third wave, presents a unique puzzle in the case of Central Asian women. Due to the constitutionally guaranteed “equal rights to work and payment for work to men and women” under the USSR (Khitarishvili 2018), which later impacted post-Soviet job opportunities, Western feminism is often perceived as “irrelevant” to the experiences of women in Central Asia.
This disconnect is further complicated by religious norms in post-Soviet Central Asia, where Islam is the most practiced religion. Importantly,
“it’s not militant or fundamentalist Islam which has posed the greatest threat to women but the reinstatement of cultural and religious norms within the public life which has led to a rise in conservative attitudes towards women” (Corcoran-Nantes 2005)
However, holding a view that the Western-based feminist discourse is irrelevant for Central Asian women due to the gender equality in the workforce and political representation during the USSR era deceives about the declining trends in female labor force participation in the region, contrasting with Europe’s upward trajectory (Andree et al. 2021).
This decline may be attributed to overlooked gender biases perpetuated by both pre-Soviet social contexts and recent social trends. Although the trends in female labor force participation vary among Central Asian countries due to differences in political and economic contexts, the overall trend still shows a decline. This research will focus specifically on Uzbekistan to further investigate the factors contributing to this phenomenon.
While numerous studies demonstrate the positive impact of female labor participation on female empowerment (Pambe 2013), it is essential to recognize that these two variables also exhibit a reciprocal relationship. Females who harbor strong gender biases often internalize these biases similarly to how individuals internalize social norms (Xenitidou 2014). Consequently, believing in predefined gender roles for women can lead them to accept these roles and not strive for job opportunities, perpetuating a self-reinforcing cycle. This reciprocal relationship is particularly pronounced in Central Asia, where women initially had equal access to education and work opportunities through communist governmental provisions. Therefore, it is the influence of societal norms that more prominently impacts females’ decisions to stay home, rather than a lack of access to opportunities.
The research question aims to investigate whether:
- The male leads (heads of household) possess gender biases and discourage their female leads (wives or partners) from working
OR
- The female leads possess gender biases themselves, which influences their decision to stay home.
Making this distinction is crucial for policy implications. If the majority of households with stay-at-home female leads have male leads who possess gender biases, but the females themselves do not hold such biases, then targeting the male leads of the households with awareness campaigns and education would be more effective than trying to empower women who are already willing to work.
Hypothesis
The woman’s decision not to participate in the labor market is more strongly correlated with the gender biases held by the male lead of the household than with the biases held by the female herself.
This hypothesis is likely to be the case for the following reasons: - Male leads of the household may possess more threat or coercive power. Even if a woman does not possess significant gender biases herself, being threatened or pressured by the male lead of the household may lead her to comply with his wishes and stay home. - The male lead’s views on gender roles can have a strong impact on the female’s self-perception. For example, a woman may enter a marriage believing that “females should work,” but if she marries a man who strongly believes that “females should stay home,” his views may influence her own beliefs and behavior over time, leading her to stay home. - The male lead’s gender biases may manifest in the form of controlling behavior, such as restricting the female lead’s access to training and/or job opportunities. This can limit the female lead’s ability to participate in the labor market, even if she does not personally hold strong gender biases.
Research Design
Data
The dataset being used for the analysis is the Baseline Project Monitoring Survey for the Rural Infrastructure Development Project (RIDP) in Uzbekistan, provided by DevLab@Penn. The survey was conducted by Al Mar Consulting with support from the RIDP Project Implementation Unit (PIU) and the World Bank. The sample includes 4,000 household interviews across 100 non-project makhallas (administrative units) in five regions: Andijon, Ferghana, Namangan, Jizzakh, and Syrdarya.
The sample of makhallas for each region and district was selected using the probability proportional to size (PPS) method. The 100 non-project makhallas were chosen from a total of 639 non-project makhallas provided in lists by the PIU.
The potential limitations of the dataset are the following: - Geographic coverage: The survey only covers five regions of Uzbekistan involved in the RIDP project, which means that the findings may not be nationally representative. The survey sample only represents the rural/village population in these regions, so the results might not apply to the entire population of Uzbekistan, particularly those living in urban areas or other regions not included in the project. - Sampling of makhallas: Only non-project makhallas were purposively sampled from lists provided by the PIU. The makhallas that were involved in the project were excluded from the household survey sample, which might introduce some selection bias.
Independent variable: The level of gender bias
To measure respondents’ gender biases, I introduced an z-score index based on their agreement levels with specific statements. These statements encompass various aspects of gender perceptions and attitudes. The respondents were asked to say to what extend they agree with the following statements:
- I3. “A university education is more important for a boy than for a girl.”
- I4. “In general, men are better political leaders than women.”
- I5. “When jobs are scarce, men should have more right to a job than women.”
- I6. “Women’s priorities should be equally valued at village meetings.”
- I7. “Generally, in this village, a wife must obey her husband.”
Additionally, the respondents were asked whether they think that hitting the wife is justified in the following scenarios:
- I9 “If she goes out without telling him”
- I10. “If she neglects the children”
- I11. “If she argues with him?”
Responses were quantified on a scale from 1 to 5, representing the level of agreement/disagreement with each statement:
- 1: Strongly disagree
- 2: Somewhat disagree
- 3: Neither disagree nor agree
- 4: Somewhat agree
- 5: Strongly agree
For questions I3, I4, I5 and I7, a higher score indicates stronger gender bias, whereas for question I6, a lower score indicates weaker gender bias. To ensure consistency across responses, I adjusted the scoring for question I6 so that a higher score correlates with stronger gender bias.
Questions I9, I10, and I11, pertaining to acceptance of gender-based violence, underwent a distinct treatment due to their binary nature. Originally, the dataset represented “Yes” as 2 and “No” as 1. Given that a “Yes” response indicates acceptance of violence, it signifies stronger gender bias. Consequently, “Yes” was assigned a score of 1, representing stronger gender bias, while “No” responses were assigned a score of 0, indicating weaker gender bias. This approach aligns with the principle applied to other responses, where higher scores correlate with greater bias and lower scores with lesser bias.
The dataset was also grouped based on sex of the respondents. Male reponsdents were grouped in one dataset, while female respondents were group into another
| Standardized Gender Bias Score | ||
| Summary Statistics | ||
| Category | Men | Women |
|---|---|---|
| Min | -1.894656 | -1.768562 |
| 1st Qu. | -0.276956 | -0.292819 |
| Median | 0.002997 | 0.002997 |
| Mean | 0.001043 | 0.001592 |
| 3rd Qu. | 0.280668 | 0.346744 |
| Max. | 1.435822 | 1.338920 |
| NA's | 4.000000 | 4.000000 |
| Data Source: DevLab@Penn (2021) | ||
Dependent variable: Female labor participation
The dependent variable is defined as a binary indicator of whether the woman in the household is employed.
For male respondents, the “work_s” column is examined, which represents the employment status of their spouse. This allows for the assessment of how the husband’s gender biases may influence their wife’s employment status. On the other hand, for female respondents, the “work” column is directly observed to determine their own employment status.
| Women's Employment Status Reported by Themselves and their Spouses | ||
| Summary Statistics | ||
| Category | Men | Women |
|---|---|---|
| Min | 0.0000 | 0.0000 |
| 1st Qu. | 0.0000 | 0.0000 |
| Median | 0.0000 | 0.0000 |
| Mean | 0.2881 | 0.2881 |
| 3rd Qu. | 1.0000 | 1.0000 |
| Max. | 1.0000 | 1.0000 |
| NA's | 8.0000 | 4.0000 |
| Data Source: DevLab@Penn (2021) | ||
Regression Model and Covariates
It is crucial to consider two important covariates when examining the relationship between gender biases and female labor force participation:
Age: Younger individuals are more likely to hold less gender-biased attitudes compared to older generations, as societal norms and values have evolved over time. However, it is also important to recognize that younger people, due to their relative lack of experience, may face more challenges in the employment market. Additionally, if younger women are married to men with gender biases, since they are likely in the early stages of their marriages, the biases coming from their husbands may not have had sufficient time to become internalized and significantly influence the women’s decisions regarding labor force participation.
Number of children: As the number of children increases, women may be more likely to opt out of the workforce to fulfill caregiving responsibilities. Additionally, having more children might reinforce gender biases regarding a woman’s role in the household, as traditional gender norms often associate women with primary caregiving duties.
To test the hypothesis, the following regression model will be used separately for female respondents and male respondents.
\[ \text{logit}(p) = \beta_0 + \beta_1 x_1 + \beta_2 x_2 + \beta_3 x_3 \]
For female respondents:
\(p\) is the probability that the respondent is employed - \(\beta\) is the coefficient for each respective variable
\(x_1\) is the gender bias score of the respondent
\(x_2\) is the age covariate
\(x_3\) is the number of children covariate
For male respondents:
\(p\) is the probability that the respondent’s spouse is employed
\(\beta\) is the coefficient for each respective variable
\(x_1\) is the gender bias score of the respondent
\(x_2\) is the age covariate
\(x_3\) is the number of children covariate
Potential Unobserved Confounders
Economic situation in the country: The overall economic conditions in Uzbekistan may have a significant impact on employment opportunities for both men and women. If the country is experiencing economic challenges or a recession, the lack of employment opportunities for women might not be solely attributable to gender biases but rather a reflection of the broader economic landscape. In such a scenario, the limited job market could be affecting all individuals
Household income and wealth: The overall financial status of the household may influence both gender biases and female employment decisions. Higher household income and wealth could reduce the pressure from women to participate in the labor force. Conversely, financial constraints may necessitate female employment, regardless of the presence of gender biases.
Empirical Extension
One important empirical extension to consider is restricting the analysis to women below the retirement age. In Uzbekistan, the retirement age for women is 55 years old (Daryo 2023). Consequently, women aged 55 and above are more likely to be out of the labor force due to retirement rather than due to the influence of gender biases. Including these retired women in the analysis could potentially confound the relationship between gender biases and female employment outcomes.
To address this issue, it would be appropriate to limit the sample to women below the age of 55. However, a challenge arises when considering the male respondents, as the dataset does not provide information on the age of their spouses. In the absence of this information, a reasonable assumption can be made that the majority of men are married to women who are younger than them. Based on this assumption, the analysis can be further refined by restricting the male respondents to those aged 60 and below. This approach aims to capture a sample where both the male respondents and their spouses are below the retirement age, minimizing the potential confounding effect of retirement on female employment status.
Findings
Interpreting the Graph
The box plot provides a comparison of gender bias scores across different employment categories. The lowest median gender bias score is observed among employed women, as indicated by the “Female Index - Employed” box. This suggests that women who are actively participating in the workforce tend to have lower levels of gender bias compared to the other groups. Conversely, the highest median gender bias score is found among unemployed women, represented by the “Female Index - Unemployed” box.
Comparing the box plots for female and male respondents reveals an interesting pattern that contradicts the initial hypothesis. The difference in gender bias levels between employed and unemployed women is more pronounced than the difference between men whose spouses are employed versus unemployed. This goes against the expectation that the influence of gender bias among men would have a more significant impact on their spouses’ employment status compared to the influence of bias among women.
The “Male Index - Female Unemployed” box, which represents the gender bias scores of men whose spouses are unemployed, shows a relatively symmetric distribution. The median line is positioned near the center of the box, indicating that an unemployed woman is almost equally likely to have a husband with stronger or weaker gender biases. This finding also challenges the hypothesis, as it suggests that the employment status of women is not strongly determined by the level of gender bias held by their husbands.
Furthermore, the “Male Index - Female Employed” box, representing the gender bias scores of men whose spouses are employed, overlaps considerably with the “Male Index - Female Unemployed” box. This implies that the employment status of women is not strongly associated with the gender bias levels of their husbands, as men with employed and unemployed spouses exhibit similar distributions of gender bias scores.
Regression Model Summary - Male Respondents
| Bivariate | Multivariate | |
|---|---|---|
| (Intercept) | 0.288*** (0.011) | 0.733*** (0.044) |
| gb_score | -0.045* (0.020) | -0.065** (0.020) |
| age | -0.008*** (0.001) | |
| children | -0.020** (0.007) | |
| Num.Obs. | 1826 | 1826 |
| R2 Adj. | 0.002 | 0.056 |
Regression Model Summary - Female Respondents
| Bivariate | Multivariate | |
|---|---|---|
| (Intercept) | 0.631*** (0.012) | 1.422*** (0.048) |
| gb_score | -0.064** (0.023) | -0.017 (0.021) |
| age | -0.017*** (0.001) | |
| children | -0.014+ (0.008) | |
| Num.Obs. | 1537 | 1537 |
| R2 Adj. | 0.004 | 0.180 |
Interpreting Regression Model Summary
The regression model provides evidence that female gender biases have a greater impact on female labor force participation compared to the influence of male gender biases on the labor force participation of their spouses. This finding contradicts the initial hypothesis, which predicted that male biases would have a more significant effect on their spouses’ employment than female biases on their own employment.
The strength of the relationship between gender biases and labor force participation can be assessed using the adjusted R-squared ($R^2$) values from the regression models. The adjusted $R^2$ value for the female respondents’ model is higher than that of the male respondents’ model (0.010 > 0.005). This indicates that the model incorporating female gender biases explains a larger proportion of the variation in female labor force participation compared to the model incorporating male gender biases and their spouses’ labor force participation. Furthermore, the statistical significance of the regression coefficients can be evaluated using the p-values. In the regression output, the p-values for the gender bias variable in both the female and male respondents’ models are significant.
However, it is important to note that while the results are statistically significant, the overall explanatory power of the models is relatively low. The adjusted $R^2$ values indicate that gender biases alone account for only a small percentage of the variation in labor force participation (18% for female respondents and 5.6% for male respondents). This implies that there are other important factors beyond gender biases that influence women’s employment decisions.
Adding Empirical Extension
Interpreting the Graph
The graph shares many similarities with the previous one, but there are notable differences arising from the empirical extension of excluding retired women from the analysis.
The removal of outliers is one of the key changes observed in the updated graph. By eliminating retired women, who are more likely to be out of the labor force due to age rather than gender biases, the distribution of gender bias scores within each employment category has become more concentrated.
Interestingly, the difference between the “Female Index - Unemployed” and “Female Index - Employed” boxes has diminished compared to the previous graph. This can be explained by the exclusion of retired women, who were more likely to be both unemployed and holding stronger gender biases. The removal of these individuals has likely caused the “Female Index - Unemployed” box to shift upward, bringing it closer to the “Female Index - Employed” box.
Another intriguing observation emerges from the removal of men over 60 years old from the analysis. Contrary to the expectation that older individuals might hold more traditional gender biases, the third quartile (Q3) for male respondents with unemployed spouses has shifted towards stronger gender biases. This finding challenges the assumption that gender biases necessarily increase with age and raises questions about the dynamics of gender biases across different generations.
Regression Model Summary - Male Respondents
| Bivariate | Multivariate | |
|---|---|---|
| (Intercept) | 0.379*** (0.013) | 0.344*** (0.071) |
| gb_score | -0.071** (0.025) | -0.068** (0.025) |
| age | 0.001 (0.001) | |
| children | -0.008 (0.010) | |
| Num.Obs. | 1301 | 1301 |
| R2 Adj. | 0.005 | 0.005 |
Regression Model Summary - Female Respondents
| Bivariate | Multivariate | |
|---|---|---|
| (Intercept) | 0.763*** (0.013) | 0.892*** (0.067) |
| gb_score | -0.046* (0.023) | -0.039+ (0.023) |
| age | -0.004** (0.001) | |
| children | 0.012 (0.010) | |
| Num.Obs. | 1136 | 1136 |
| R2 Adj. | 0.003 | 0.010 |
Interpreting Regression Model Summary
For male respondents, adding the covariates of number of children and age to the regression model did not substantially impact the \(R^2\) value (0.005), indicating that these variables did not significantly strengthen the relationship between the independent and dependent variables. This can be explained by the following factors:
Age: As previously mentioned, one reason why age might be a covariate is that as women grow older, they may become more entrenched in their relationships and more likely to internalize gender biases over time. However, this is often due to the influence of living with a husband who already holds gender biases. It is unlikely that a wife’s presence will significantly alter the pre-existing gender biases of a male partner, especially in male-led households. Instead, the husband’s biases may even be reinforced over time.
Number of children: In this case, the number of children may not have a substantial impact on a man’s gender biases. While some evidence suggests that larger families are associated with more conservative, traditional households, other factors could be at play. For example, the number of children may have been beyond the couple’s control, or the households in the sample may have a similar number of children, reducing the variable’s explanatory power.
In contrast, for female respondents, adding the covariates increased the \(R^2\) value from 0.003 to 0.010, indicating that age and number of children did have an impact on the variables, as outlined in the method section.
The regression analysis provides a more accurate assessment of the strength of the relationship between variables compared to the graph interpretation. Likewise the previous regression analysis, the \(R^2\) values suggest that female gender bias has a greater impact on female employment than male gender bias has on their spouses’ employment. Interestingly, without the inclusion of covariates, the conclusion would have been the opposite, as the \(R^2\) value for male respondents in the bivariate column is higher than the \(R^2\) value for female respondents.
Importantly, the empirical extension significantly decreased the \(R^2\) value for female respondents from 18% to 1%. It’s expected, since we ruled out the retired women – who probably had more gender bias due to being exposed to different culture norms and are less likely to be unemployed due to passing retirement age.
Discussion and Policy Implications
The final project aimed to investigate the impact of gender biases on female labor force participation (FLP) in Uzbekistan. The analysis focused on examining the relationship between gender bias scores and employment outcomes for both women and men, with the hypothesis that male gender biases would have a stronger influence on their spouses’ labor force participation compared to the impact of female gender biases on their own employment.
The results of the analysis rejected the hypothesis, indicating that female gender biases have a more significant impact on FLP than the gender biases of males have on their spouses’ labor participation.
Interestingly, controlling for age revealed that it played a strong role in influencing the findings. This implies that future research could benefit from exploring how gender biases have changed across generations among men and women in Uzbekistan. Investigating questions such as whether men in Central Asia are becoming more gender-biased while females are becoming less gender-biased in the region, or how gender biases among people living during the USSR differ from the contemporary generation, could provide valuable insights. Although such questions would be challenging to investigate due to the need to control for many factors, qualitative research could offer an interesting approach to empirically evaluate widespread pro-communist feminist ideals.
Another relationship that merits further exploration, but was beyond the scope of this paper, is the level of adherence to religion and its association with gender biases, especially considering the recent rise of Islamic thought in Central Asia.
It is important to acknowledge that the survey used in this analysis did not specifically focus on gender bias, which limited the number of questions related to this topic. Incorporating a more comprehensive set of questions specifically designed to capture gender biases could have allowed for a more accurate representation of each individual’s attitudes and beliefs.
In terms of policy implications, the findings suggest that addressing gender biases alone may not be sufficient to significantly increase female labor force participation in Uzbekistan. While promoting gender equality and challenging traditional gender roles is crucial, policymakers should also focus on addressing other barriers to women’s employment.